By Andrew Bartolini, Chief Research Officer, Ardent Partners
Every procurement transformation has a blind spot, and for most enterprises, it sits right at the beginning of the process. Not at sourcing. Not at contracts. Not even at purchase orders. The dysfunction begins the moment a business user decides they need something and then figures out how to get it without ever touching the procurement system.
That moment, multiplied across thousands of employees, dozens of categories, and every conceivable channel of communication (email, Teams message, hallway conversation, informal workaround), is what the market now calls the intake problem. And after years of being an afterthought, it has become one of the most strategically important problems in enterprise procurement.
I recently sat down with Bikash Mohanty, VP of Solution Consulting and Product Strategy at Zycus, for a briefing and live demonstration of Merlin Intake, the company’s AI-powered intake management solution. What follows is a summary of what Zycus showed us and what we found notable. Part Two will offer Ardent’s views on what it means, why it matters, and what procurement and AP leaders should be thinking about.
Why Intake, Why Now
To understand why intake has become such a hot category, you have to understand the full weight of what “fragmented” actually means in the context of procurement infrastructure.
Ardent Partners’ market research has consistently shown that process and technology fragmentation is one of the most persistent barriers to procurement performance. In practice, this fragmentation erodes value, but it doesn’t just create inefficiency, it creates organizational behavior. Stakeholders who find their procurement system difficult to use, learn to bypass it. They send an email. They call a colleague. They order on a corporate card and submit an expense. Each of those workarounds represents spend that procurement never touched, a supplier relationship that was never rationalized, a policy that was never applied, and a process that started broken and produced more broken outcomes downstream.
The traditional fixes like better forms, improved portals, and more training, treat this as a change management problem. It isn’t, or at least, it isn’t only that. It is also an architecture problem. Most enterprise procurement technology was designed to support the professional buyer, not the business user who buys something occasionally and has no interest in investing their time and effort in learning a new system. When the front door to procurement is hard to open, people stop using it.
The emergence of conversational AI has changed what is possible. For the first time, the interface to procurement can be as simple as saying what you need. That shift, from form-filling to natural language conversation, is the primary reason intake has moved from a niche feature to a serious strategic category in a very short time. Several solution providers have entered the space. Some are standalone. Some are embedded within broader S2P suites. That distinction, as we’ll discuss, matters considerably.
The Zycus Position: Built-In, Not Bolted On
Zycus entered this conversation with a specific and pointed argument: intake management only works well when it is native to the broader Source-to-Pay suite, not when it is a standalone tool or third-party add-on layered on top of existing infrastructure.
Our briefing began with a focus on the architecture. Merlin Intake is embedded directly within the Zycus S2P platform. It is native to the larger suite; there is no integration or any APIs. Every module in the suite shares the same data core. When a request comes in through Merlin Intake, it moves through that system without handoffs, translation layers, or integration points that can leak value or introduce delay.
This matters because the intake problem is not just a front-end UX problem. It is also a data continuity problem. A standalone intake tool that captures a request beautifully and then hands it off to five different downstream systems via five different integrations has solved the front-end experience, but recreated the fragmentation one step later in the process. Zycus’ argument is that the best fix is a unified Intake-to-Pay system where the request, the workflow, the policy enforcement, and the resulting transaction all operate on a single data model.
That argument is not unique to Zycus. As you would imagine, other S2P suite providers make a similar case. It is a legitimate one.
What Merlin Intake Actually Does
The live demo anchored the briefing with a real purchasing scenario: a requisition for catering services for a marketing event in New York, $25,000 budget, 300 attendees, with an unfamiliar supplier. He typed the request in plain language. There was no form or mandatory fields or dropdown menus.
The system identified the need for a purchase requisition via natural language processing and extracted the relevant details from the input, including the service type, the location, the delivery date, the budget, the supplier name. The system asked a few clarifying questions: how many attendees, what category code to apply (and when “the user” said he didn’t know, the system suggested one). It checked whether a supplier existed in the system (it didn’t) and identified that this request would require a supplier onboarding step before a PO could be created. The system then projected the expected PO cycle time based on the standard company process and current policy.
Interestingly, the system adjusted the total requisition amount from $25,000 to $24,900 after the attendee count was entered. It had calculated the per-head cost automatically. Now, a $100 delta on this PO is a minor thing in isolation, but it illustrates that the system is not simply collecting information, but rather is processing it in context.
The policy enforcement layer is similarly designed to be invisible to the user while doing real governance work underneath. Administrators define policies in plain English, not in code nor with drag-and-drop workflow builders. The AI interprets those policies and applies them conversationally. A user asking to purchase something that requires a sourcing event gets guided toward a sourcing event, without necessarily knowing that is what is happening. Compliance becomes a byproduct of a good user experience, not a hurdle.
Notifications and status tracking run through the same interface. A business user who submitted a requisition can ask about the status of their purchase order, query which POs are pending receipt, or request a summary of their spend. And they do so within the same conversational window where they originated the request. For organizations that have spent years managing the downstream communications that follow every purchase request, this is a meaningful improvement.
Merlin Intake operates natively within Microsoft Teams as well as through a web portal, which addresses one of the more practical adoption challenges: meeting users where they are rather than asking them to open another application.
The AI Architecture
Zycus is using ChatGPT as its underlying LLM, with a proprietary prompting architecture and approach built on top. While early versions of the system showed meaningful hallucination issues that the team had to work through over time. They have, by their account, reduced those significantly through prompt engineering and the system architecture, drawing on RAG (retrieval-augmented generation) techniques to ground responses in actual policy documents and company data, rather than relying on the LLM’s general knowledge.
This is the right approach, and the candor about the early challenges was refreshing. Every solution provider in this space is navigating the same tension between the remarkable capability of large language models and their well-documented tendency to generate plausible-sounding nonsense when operating outside their training data. The answer is not to shy away from that challenge but to engineer around it, which is what Zycus describes doing.
Zycus has placed these capabilities into production in 2026 and they could be genuinely significant: connecting the AI directly to the underlying database rather than relying on APIs to surface data. This means querying the system to surface purchase orders at risk of late delivery and getting results that draw on historical delivery data, due dates, and supplier performance context. If that capability matures and productizes, it represents a new and impressive category of intelligence.
The Customers and the Use Cases
Zycus named two customers in the context of the briefing, but we will keep them anonymous. One is using Merlin Intake to orchestrate 17 different procurement workflows. This scope gives a sense of the configurability required to serve a large, complex organization. The second customer has multiple procurement workflows connected to the S2P system, including travel and expense integrations.
We did not speak to either company, but the use case coverage described by Zycus is broad: buy journeys covering catalog and non-catalog purchases, hardware, software, and services; contracting journeys for renewals and amendments; invoice journeys covering both PO and non-PO scenarios; and a range of edge cases including NDAs, vendor termination letters, staffing requests, and event management. That breadth reflects a deliberate positioning of Merlin Intake as a true front door to procurement that can operate as control tower for every procurement-adjacent request that enters the organization.
Watch for Part 2 of this series, coming soon.
RELATED RESOURCE
Returning to Zycus Horizon, Building the Case for Agentic AI in Procurement
